This document summarizes analyses to optimize placement of Ebola treatment units (ETUs) in Liberia. Models were developed to forecast Ebola incidence at the county level and predict spatial disease burden. Various allocation strategies were evaluated, including placements based on population and predicted burden. The analyses compared two optimization methods and evaluated network reliability issues. Future work proposed iterative planning, mini-ETUs, alternative optimization objectives, and using updated data to refine recommended locations.
Biopesticide (2).pptx .This slides helps to know the different types of biop...
Modeling the Ebola Outbreak in West Africa, October 15th 2014 update
1. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Op2mal
Ebola
Treatment
Unit
Placement
in
Liberia
Oct
15th
Update
Bryan
Lewis
PhD,
MPH
(blewis@vbi.vt.edu)
Caitlin
Rivers
MPH,
Eric
Lofgren
PhD,
James
Schli,,
Alex
Telionis
MPH,
Henning
Mortveit
PhD,
Dawen
Xie
MS,
Samarth
Swarup
PhD,
Hannah
Chungbaek,
Keith
Bisset
PhD,
Maleq
Khan
PhD,
Chris
Kuhlman
PhD,
Farzaneh
Tabataba,
Anil
Vullikan2,
Dana
Kuan
(DTRA)
Stephen
Eubank
PhD,
Madhav
Marathe
PhD,
and
Chris
Barre.
PhD
Technical
Report
#14-‐111
2. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Features
of
refined
analysis
• Added
ODE
model
based
burden
predic2ons
• Added
travel
speeds
to
network
• Because
outputs
of
prior
work
all
similar:
– Focused
exclusively
on
“Pa2ent
Direct
to
ETU”
and
the
LandScan™
Grid
for
both
methods
• Outputs
include:
– Alloca2on
for
en2re
na2on
based
on
Popula2on
(12
new
centers)
– Alloca2on
for
northern
coun2es
based
on
Ebola
Burden
(6
centers)
– Alloca2on
for
en2re
na2on
based
on
Ebola
Burden
(12
centers)
Ignoring
2
new
centers
already
planned
for
Monrovia
and
Kakata
3. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
ODE
Model
to
forecast
incidence
• Fit
SEIR
models
to
Liberian
coun2es
with
>30
cases
and
>10
new
cases
in
the
last
21
days
• Forecasted
to
December
1st,
2014
4. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
ODE
models
used
to
forecast
incidence
in
4
coun2es
5. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Predicted
Spa2al
Burden
6. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Popula2on
Based
Alloca2on
7. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Burden
Based
Alloca2on
(6)
8. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Burden
Based
Alloca2on
(12)
9. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Alloca2on
based
on
an
alterna2ve
method:
MapOp2mizer
10. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
MapOp2mizer
11. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Comparison
of
Two
Methods
12. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Comparison
of
Two
Methods
13. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Future
Work
• Network
reliability!
– Even
main
roads
can
be
washed
out
or
impassable.
• Place
Mini-‐ETUs
– 10-‐20
bed
facili2es
placed
between
main
ETUs
Maryland
Avenue
from
Pleebo
to
Harper
(from
John
Etherton).
14. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Future
Work
• Itera2ve
Planning
approach
– Provide
candidate
loca2ons,
model
with
these
– As
on
the
ground
data
is
provided
ruling
out
different
sites,
readjust
and
provide
the
next
“op2mal”
solu2on
• Try
other
Op2mal
alloca2ons:
Maximum
A.endance
versus
K-‐medians
– K-‐medians
is
most
“equitable”
solu2on
– 1
person
at
100
miles
=
100
people
at
1
mile
– MA
ignores
those
beyond
distance
threshold
• Pro:
maximizes
availability
in
high
density
areas
• Con:
ignores
very
remote
popula2on
centers
15. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Review
of
briefing
on
October
7th
15
16. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Features
and
assump2ons
• Op2mized
Loca2ons
for
Southeast
Liberia
Only
– Grand
Gedeh,
Grand
Kru,
River
Cess,
River
Gee,
Maryland,
and
Sinoe
Coun2es
• Delivered
report
to
DTRA
on
2014-‐10-‐06
• Limita2ons:
– All
roads
and
rivers
weighted
equally
– Ebola
case
load
not
used
– Did
not
include
network
reliability
17. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Compe2ng
Methods
• Loca2on-‐Alloca2on
– Run
in
Esri®
ArcGIS™
10.1
SP1
Network
Analyst
– Solves
k-‐medians
problem:
places
facili2es
to
minimize
weighted
travel
2me
for
all
people
• MapOp2mizer
– Wri.en
in
Python
using
NetworkX1
Library
– Solves
via
Dijkstra’s
Algorithm2
18. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Data
Sources
• Road
Network:
The
Liberia
Ins2tute
of
Sta2s2cs
and
Geo-‐Informa2on
(LISGIS)
3
– Shapefile
from
John
Etherton
(personal
communica2on)
• River
Network:
Diva-‐GIS4
• OpenStreetMap:
Very
detailed,
but
network
is
disconnected,
and
missing
important
rural
roads
• Popula2on:
LandScan5
or
WorldPop6
21. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Prior
Work
• Eight
runs
of
LocAll
– WorldPop
or
LandScan
™
popula2on
grids
– “Pa2ent
Direct
to
ETU”
or
“Pa2ent
to
Clinic
to
ETU”
– Six
or
Seven
ETUs
placed
– All
outputs
very
similar
• Two
runs
of
MapOp2mizer
– LandScan
+
Direct
– Six
or
Seven
ETUs
placed
22. DRAFT
–
Not
for
a.ribu2on
or
distribu2on
Sources
1. Hagberg,
A.,
Swart,
P.,
&
S
Chult,
D.
(2008).
Exploring
network
structure,
dynamics,
and
func2on
using
NetworkX
(No.
LA-‐UR-‐08-‐05495;
LA-‐
UR-‐08-‐5495).
Los
Alamos
Na2onal
Laboratory
(LANL).
2. Dijkstra,
E.
W.
(1959).
A
note
on
two
problems
in
connexion
with
graphs.
Numerische
mathema2k,
1(1),
269-‐271.
3. Etherton,
John
(2014).
[Personal
Communica2on
(2014-‐09-‐18)].
4. Hijmans,
RJ,
Guarino,
L,
Bussink,
C,
Mathur,
P,
Cruz,
M,
Barrentes,
I,
&
Rojas,
E.
(2004).
DIVA-‐GIS.
Vsn.
5.0.
A
geographic
informa2on
system
for
the
analysis
of
species
distribu2on
data.
Manual
available
at
h.p://
www.diva-‐gis.org.
5. Bright,
Eddie
A.,
Coleman,
Phil
R.,
Rose,
Amy
N.,
&
Urban,
Marie
L.
(2014).
LandScan
2013.
In
LLC
UTBa.elle
(Ed.),
(2013
ed.).
Oak
Ridge,
TN:
Oak
Ridge
Na2onal
Laboratory.
6. Tatem,
A.J.,
Gething,
P.W.,
Bha.,
S.,
Weiss,
D.,
&
Pezzulo,
C.
(2014).
WorldPop
2014:
Pilot
high
resolu2on
poverty
maps:
University
of
Southampton
/
Oxford.